#Load packages
pacman::p_load(plotly, ggstatsplot, knitr, patchwork, ggdist, ggthemes, tidyverse)
#Import data
flatprice <- read_csv("data/resale-flat-prices-based-on-registration-date-from-jan-2017-onwards.csv", show_col_types = FALSE)Take-home Exercise 3
Visual Analytics of Resale Prices of Singapore Public Housing Properties
1. Overview
This exercise aims to uncover the salient patterns of the resale prices of public housing property by residential towns and estates in Singapore using appropriate analytical visualisation techniques. The visualization is designed using ggplot2, its extensions, and tidyverse packages.
The original dataset was downloaded from Data.gov.sg titled Resale flat princes based on registration date from Jan-2017 onwards.
The file downloaded was resale-flat-prices-based-on-registration-date-from-jan-2017-onwards.csv
The focus of the study is on 3-ROOM, 4-ROOM and 5-ROOM types for 2022 period.
2. Data Preparation
2.1 Install R packages and import dataset
The code chunk below uses pacman::p_load() to check if packages are installed. If they are, they will be launched into R. The packages installed are
plotly: Used for creating interactive web-based graphs.ggstatsplot: Used for creating graphics with details from statistical tests.knitr: Used for dynamic report generationpacthwork: Used to combine plotsggdist: Used for visualising distribution and uncertaintyggthemes: Provide additional themes forggplot2tidyverse: A collection of core packages designed for data science, used extensively for data preparation and wrangling.All packages can be found within CRAN.
Import data from csv using readr::read_csv() and store it in variable flatprice.
2.2 Data wrangling
Looking at the data below, we notice few problems
month is in
<chr>format (“yyyy-mm”), which is not very useful for filtering for 2022 periodlease_commence_date is in
<dbl>format. It needs to be converted to<chr>remaining_lease is in
<chr>format. It needs to be reformatted to<dbl>in years
flatprice# A tibble: 146,215 × 11
month town flat_…¹ block stree…² store…³ floor…⁴ flat_…⁵ lease…⁶ remai…⁷
<chr> <chr> <chr> <chr> <chr> <chr> <dbl> <chr> <dbl> <chr>
1 2017-01 ANG MO… 2 ROOM 406 ANG MO… 10 TO … 44 Improv… 1979 61 yea…
2 2017-01 ANG MO… 3 ROOM 108 ANG MO… 01 TO … 67 New Ge… 1978 60 yea…
3 2017-01 ANG MO… 3 ROOM 602 ANG MO… 01 TO … 67 New Ge… 1980 62 yea…
4 2017-01 ANG MO… 3 ROOM 465 ANG MO… 04 TO … 68 New Ge… 1980 62 yea…
5 2017-01 ANG MO… 3 ROOM 601 ANG MO… 01 TO … 67 New Ge… 1980 62 yea…
6 2017-01 ANG MO… 3 ROOM 150 ANG MO… 01 TO … 68 New Ge… 1981 63 yea…
7 2017-01 ANG MO… 3 ROOM 447 ANG MO… 04 TO … 68 New Ge… 1979 61 yea…
8 2017-01 ANG MO… 3 ROOM 218 ANG MO… 04 TO … 67 New Ge… 1976 58 yea…
9 2017-01 ANG MO… 3 ROOM 447 ANG MO… 04 TO … 68 New Ge… 1979 61 yea…
10 2017-01 ANG MO… 3 ROOM 571 ANG MO… 01 TO … 67 New Ge… 1979 61 yea…
# … with 146,205 more rows, 1 more variable: resale_price <dbl>, and
# abbreviated variable names ¹flat_type, ²street_name, ³storey_range,
# ⁴floor_area_sqm, ⁵flat_model, ⁶lease_commence_date, ⁷remaining_leaseis.na() function is also used to confirm that there are no missing values in the flatprice dataset.
#Check for missing values
any(is.na(flatprice))[1] FALSEThe code chunk below performs the required data wrangling to clean flatprice dataset and store it in new variable flatpriceclean.
- Filter flat_type for 3 ROOM, 4 ROOM, and 5 ROOM as this is the scope of the study using
dplyr::filter() - Convert the month variable to date using
as.Date(). Store the year and month to the respective new variables year and month usingdplyr:mutate(). They can then be converted to string (discrete variables) usingas.character(). Afterwards, we can filter the year variable to 2022 usingdplyr::filter(), which is the scope of the study - Extract the year and month digits from remaining_lease variable using
str_extract()function. Sum the year digit and (month digit/12) to obtain the years of remaining lease and convert it to<dbl>format usingas.numeric(). They are then rounded to 1 decimal place usinground(). Assign new variable called remaining_lease_years usingdplyr::mutate() - Create new variable called resale_price_persqm to divide the resale_price with floor_area_sqm. This is performed to normalize the resale price to flat area. They are then rounded to 1 decimal place using
round().The new variable is assigned usingdplyr::mutate - Convert lease_commence_date to
<int>usingas.integer()
#Data preparation
#store the new dataset in new variable flatpriceclean
flatpriceclean <- flatprice |>
#filter for 3-ROOM, 4-ROOM, 5-ROOM
filter(flat_type %in% c('3 ROOM','4 ROOM','5 ROOM')) |>
#reformat month and split it to month and year. Use year to filter for 2022
mutate(year = as.character(format(as.Date(paste(month, "-01", sep="")), "%Y")),
month = as.character(format(as.Date(paste(month, "-01", sep="")), "%m")),
.before = 1)|>
filter(year == 2022) |>
#mutate remaining_lease to remaining_lease_years
mutate(remaining_lease_years = round((as.numeric(str_extract(remaining_lease, "^[0-9]+")) +
ifelse(is.na(as.numeric(str_extract(remaining_lease, " [0-9]+"))), 0, as.numeric(str_extract(remaining_lease, " [0-9]+")))/12), digits = 1),
#create new variable called resale_price_persqm
resale_price_persqm = round(resale_price/floor_area_sqm, digits = 1),
#convert lease_commence_date to integer
lease_commence_date = as.integer(lease_commence_date),
.after = remaining_lease) The final dataset flatpriceclean is displayed below.
kable(head(flatpriceclean), "simple")| year | month | town | flat_type | block | street_name | storey_range | floor_area_sqm | flat_model | lease_commence_date | remaining_lease | remaining_lease_years | resale_price_persqm | resale_price |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| 2022 | 01 | ANG MO KIO | 3 ROOM | 320 | ANG MO KIO AVE 1 | 07 TO 09 | 73 | New Generation | 1977 | 54 years 05 months | 54.4 | 4904.1 | 358000 |
| 2022 | 01 | ANG MO KIO | 3 ROOM | 225 | ANG MO KIO AVE 1 | 07 TO 09 | 67 | New Generation | 1978 | 55 years 01 month | 55.1 | 5298.5 | 355000 |
| 2022 | 01 | ANG MO KIO | 3 ROOM | 331 | ANG MO KIO AVE 1 | 07 TO 09 | 68 | New Generation | 1981 | 58 years | 58.0 | 4970.6 | 338000 |
| 2022 | 01 | ANG MO KIO | 3 ROOM | 534 | ANG MO KIO AVE 10 | 07 TO 09 | 82 | New Generation | 1980 | 57 years 02 months | 57.2 | 5122.0 | 420000 |
| 2022 | 01 | ANG MO KIO | 3 ROOM | 578 | ANG MO KIO AVE 10 | 04 TO 06 | 67 | New Generation | 1980 | 57 years 01 month | 57.1 | 4895.5 | 328000 |
| 2022 | 01 | ANG MO KIO | 3 ROOM | 452 | ANG MO KIO AVE 10 | 01 TO 03 | 83 | New Generation | 1979 | 56 years 07 months | 56.6 | 4337.3 | 360000 |
3. Visualisation
3.1 Exploratory Data Visualisation
The plots here are preliminary in nature but designed with interactivity to allow users to perform Exploratory Data Analysis (EDA) Visualisation to study the data. Hence plotly is mainly used.
3.1.1. Interactive scatterplot
Design Consideration
Scatterplots are generally used to discover relationship between two continuous variables. As such, the visualization below allows users to select the x-axis and y-axis of the continuous variables they wish to study. Considerations :
Dropdown menu to select variables for y-axis and x-axis
y-axis selection is resale_price and resale_price_persqm. This is aligned with the study purpose of discovering patterns of resale price. This allows users to also see the intent of normalising resale_price by floor_area_sqm
x-axis selection is other continuous variables, namely: remaining_lease_years, lease_commence_date, and floor_area_sqm
As the plots are expected to be very cluttered, opacity is introduced with white border
Tooltip indicating variables of interest : remaining_lease_years, floor_area_sqm, and resale_price_persqm
Preparation of visualisation
plot_ly is used to prepare the interactive plot. Steps taken are
Initiating base scatterplot, indicated by
type = 'scatter'.markerargument is used to introduceopacityandline(white plot border)Tooltip customisation is created using
hovertemplateargumentlayoutargument is used to add plot title, x-axis title, and y-axis titleTo create the dropdown menu for parameters of x-axis and y-axis, the
updatemenusargument is used to create respectivebuttons
Refer to code below for more details
Show the code
#Initiating the base plot
plot_ly(data = flatpriceclean,
x = ~remaining_lease_years,
y = ~resale_price_persqm,
hovertemplate = ~paste("<br>Resale Price per sqm:", resale_price_persqm,
"<br>Floor Area (sqm):", floor_area_sqm,
"<br>Remaining Lease (Year):", remaining_lease_years),
type = 'scatter',
mode = 'markers',
marker = list(opacity = 0.6,
sizemode = 'diameter',
line = list(width = 0.2, color = 'white'))) |>
#Generating plot, x-axis, and y-axis title
layout(title = "Interactive scatterplot of resale price vs selected factors\nResale transactions, 2022",
xaxis = list(title = "Remaining Lease (Year)"),
yaxis = list(title = "Resale Price per sqm (SGD)"),
#creating dropwdown menus to allow selection of parameters on x-axis and y-axis
updatemenus = list(list(type = "dropdown",
direction = "up",
xref = "paper",
yref = "paper",
xanchor = "left",
yanchor = "top",
x = 1,
y = 0,
buttons = list(
list(method = "update",
args = list(list(x = list(flatpriceclean$remaining_lease_years)),
list(xaxis = list(title = "Remaining Lease (Year)"))),
label = "Remaining Lease"),
list(method = "update",
args = list(list(x = list(flatpriceclean$lease_commence_date)),
list(xaxis = list(title = "Year of Lease Commenced"))),
label = "Lease Commenced"),
list(method = "update",
args = list(list(x = list(flatpriceclean$floor_area_sqm)),
list(xaxis = list(title = "Floor Area (sqm)"))),
label = "Floor Area")
)
),
list(type = "dropdown",
xref = "paper",
yref = "paper",
xanchor = "left",
yanchor = "top",
x = 0.04,
y = 0.95,
buttons = list(
list(method = "update",
args = list(list(y = list(flatpriceclean$resale_price_persqm)),
list(yaxis = list(title = "Resale Price per sqm (SGD)"))),
label = "Resale Price/Area"),
list(method = "update",
args = list(list(y = list(flatpriceclean$resale_price)),
list(yaxis = list(title = "Resale Price (SGD)"))),
label = "Resale Price")
)
)
)
)Insights
The most obvious pattern is the direct positive correlation between resale price and floor area. Hence the decision to normalize the resale price is justified. From here onwards, we will focus more on resale price per sqm.
Plotting resale price per sqm vs remaining lease reveals somewhat positive exponential correlation. There seems to sharper increase in gradient if the remaining lease is beyond 80 years
As expected, the same pattern is revealed when resale price per sqm is plotted against the year of lease commenced. In fact the year of lease commenced and remaining lease are directly correlated to each other as shown in the plot below. Hence it makes sense to use one of them in future analysis. In this case, we will only use remaining lease as it is more intuitive.
Show the code
ggplot(data = flatpriceclean)+
geom_point(aes(x = lease_commence_date,
y = remaining_lease_years)) +
labs(
x = "Year of Lease Commenced",
y = "Remaining Lease\n(Year)") +
theme(axis.title.y = element_text(angle = 0))
3.1.2. Interactive boxviolin plot
Design Consideration
In contrast to scatterplots, boxplots are generally used to discover relationship between continuous variables and discrete variables. Violin plots can also be added to visualize the kernel density of the data.
As such, the visualization below allows users to select the x-axis of the discrete variables they wish to study. Considerations :
- Dropdown menu to select variables for x-axis
- y-axis selected is resale_price_persqm. This is aligned with the study purpose of discovering patterns of resale price
- x-axis selection is the discrete variables, namely: flat_type, flat_model, storey_range, town, and month. Note that month (transaction month) is treated as discrete variable as we want to find out any seasonal variations (statistically significant difference in mean/median) of price.
- x-axis title is omitted as it will be displayed by the selected dropdown menu
- Combining boxplots and violin plots to visualize the quartiles and kernel density of the distribution
- Tooltip to show mean, quartiles, and outliers
- Plot can be zoomed-in or out as it might be difficult to get details with large number of categories (i.e., town)
- Order the variables by ascending order (mean) if they are not lexicographically important (i.e., flat_type, storey_range, and month
Preparation of visualisation
plot_ly is used to prepare the interactive plot. Steps taken are
Initiating base scatterplot, indicated by
type = 'violin'. Noteboxvisibleargument is set toTto ensure the boxplot is visible. The same is applied tomeanlinevisibleto ensure meanline is visible.layoutargument is used to add plot title, and y-axis titleTo create the dropdown menu for parameters of x-axis, the
updatemenusargument is used to create respectivebuttonscategoryorderis set for each button to determine the order of discrete variables.category ascendingwill ensure lexicographic order whilemean ascendingwill ensure mean order.
Refer to code below for more details
Show the code
#Initiating the base plot
plot_ly(data = flatpriceclean,
x = ~flat_type,
y = ~resale_price_persqm,
type = "violin",
marker = list(line = list(width = 2)),
box = list(visible = T),
meanline = list(visible = T)) |>
#Generating plot and y-axis title
layout(title = "Distribution of resale price by selected factors \nResale transactions, 2022",
xaxis = list(title = ""),
yaxis = list(title = "Resale Price per sqm (SGD)"),
#creating dropwdown menus to allow selection of parameters on x-axis
updatemenus = list(list(type = 'dropdown',
xref = "paper",
yref = "paper",
xanchor = "left",
x = 0.04,
y = 0.95,
buttons = list(
list(method = "update",
args = list(list(x = list(flatpriceclean$flat_type)),
list(xaxis = list(categoryorder = "category ascending"))),
label = "Flat Type"),
list(method = "update",
args = list(list(x = list(flatpriceclean$flat_model)),
list(xaxis = list(categoryorder = "mean ascending"))),
label = "Flat Model"),
list(method = "update",
args = list(list(x = list(flatpriceclean$storey_range)),
list(xaxis = list(categoryorder = "category ascending"))),
label = "Storey Height"),
list(method = "update",
args = list(list(x = list(flatpriceclean$town)),
list(xaxis = list(categoryorder = "mean ascending"))),
label = "Town"),
list(method = "update",
args = list(list(x = list(flatpriceclean$month)),
list(xaxis = list(tickmode = "array")),
list(color = list(flatpriceclean$month))),
label = "Transaction Month")
)
)
)
)Insights
Note that insights can be garnered by zooming in some plots with high number of categories
When plotted by flat_type, preliminary EDA reveals minimal difference between the means/medians. This is expected since we are using resale_price_persqm. Note that flats with more rooms tend to be bigger in area, hence when normalized by area, the difference in resale price among different types of flat is minimized. It is interesting to note that 5 ROOM flat type has the lowest mean overall.
When plotted by flat_model, preliminary EDA reveals higher mean/median resale_price_persqm for models like DBSS, Terrace, Premium, S2 and S1. However, it is important to note that these types tend to be few in the market as well, which likely forms the more exclusive models. Note large variance are observed for improved and Model A types.
When plotted by storey_range, preliminary EDA reveals higher mean/median resale_price_persqm for higher storey. Note large variance are observed for 19 to 27th storey units.
When plotted by town, preliminary EDA reveals differences in mean/median resale_price_persqm for different town estates. At the lower end, we can find newly developed or further towns (i.e., Woodlands, Pasir Ris, Jurong West), while at the opposite end, we find more mature or centrally located towns (i.e., Central Area, Queenstown, Kallang/Whampoa). Note that the data looks skewed right with most of the mean is higher than the median and high presence of outlier on the higher end.
Preliminary EDA reveals minimal difference in resale_price_persqm between transaction months.
However, these observations need to be confirmed using appropriate hypothesis techniques. Since the focus of the study is on resale price by residential towns/estates, only this will be considered.
3.2 Confirmatory Data Analysis Visualisation
The plots here are focused in visual statistical testing used in Confirmatory Data Analysis. Hence ggstatsplot is mainly used.
Design Consideration
To ensure that best practices (APA standard) of statistical reporting is followed, ggstatsplot library is used in this section. Considerations :
As revealed by EDA, town variables have 26 categories. Hence
ggbetweenstatswill be used for one-way ANOVA testing.Median will be used instead of mean as point estimates due to high presence of outliers and skewness of data.
Since we are unable to confirm normality assumption, non-parametric test (Kruskal-Wallis) will be used for hypothesis testing. Note that the null hypothesis is no difference between median resale price per area by towns/estates.
Given the large number of categories, two plots are considered, splitting the towns seeing >= 850 transactions and those seeing <850 transactions in 2022. This is mainly to ensure that the selected towns can fit in single plot (13 towns each) as having 26 towns in one single plot causes it to be too cluttered. This also helps to reduce bias against variations in number of observations. Note that the y-axis range needs to be fixed for the two plots for easier comparison.
Preparation of visualisation
ggstatsplot is used to prepare the plot. Steps taken are
Order the towns by the median of resale_price_persqm in ascending order using
forcats::fct_reorder. This will ensure the towns are ordered accordingly for easier observation.ggstatsplot::ggbetweenstats()is used to display the one-way ANOVA plot. Note that the dataset is grouped by town and filtered when count >=850. This is achieved usingdplyr::group_by()anddplyr::filter()As above design consideration, non-parametric test is considered hence
type = "np"Comparisons are made pairwise between non-significant pairs as they are the points of interest. Note that the majority of the pairs comparison are significant as well, hence it will clutter the plots unnecessarily. The argument
pairwise.display = "ns"Two plots are created and assigned to
p1andp2respectivelyTwo sets of color are manually provided using
scale_color_manualofggplot2. This is because the basic palette is not able to support large number of colors (26)scale_y_continuousis used to set the y-axis range between 4,000 and 20,000patchworkis added to combine the two plots using operator/plot_annotation()is used to configure the title, subtitle and caption of the combined visualisation
Refer to code below for more details
Show the code
#Order the town variables by median of resale price
flatpriceclean$town <- fct_reorder(flatpriceclean$town, flatpriceclean$resale_price_persqm, median)
#set the color pallete
col13_1 <- c("dodgerblue2", "#E31A1C",
"green4",
"#6A3D9A",
"#FF7F00",
"gray30", "gold1",
"skyblue2", "#FB9A99",
"palegreen2",
"#CAB2D6",
"#FDBF6F",
"gray80")
col13_2 <- c("khaki2",
"maroon", "orchid1", "deeppink1", "blue1", "steelblue4",
"darkturquoise", "green1", "yellow4", "yellow3",
"darkorange4", "brown", "gray50")
#Initiating the base plot
p1 <- ggbetweenstats(
data = flatpriceclean|>
group_by(town) |>
filter(n() >= 850),
x = town,
y = resale_price_persqm,
ylab = "Resale Price per sqm (SGD)",
title = "Plot 1. One-way ANOVA (towns >= 850 transactions)",
type = "np",
pairwise.comparisons = TRUE,
pairwise.display = "ns",
mean.ci = TRUE,
p.adjust.method = "fdr",
messages = FALSE
) +
scale_color_manual(values = col13_1) +
theme(axis.title.x = element_blank()) +
scale_y_continuous(limits = c(4000, 20000))
p2 <- ggbetweenstats(
data = flatpriceclean|>
group_by(town) |>
filter(n() < 850),
x = town,
y = resale_price_persqm,
xlab = "Planning Town",
ylab = "Resale Price per sqm (SGD)",
title = "Plot 2. One-way ANOVA (towns < 850 transactions)",
type = "np",
pairwise.comparisons = TRUE,
pairwise.display = "ns",
mean.ci = TRUE,
p.adjust.method = "fdr",
messages = FALSE
) +
scale_color_manual(values = col13_2) +
scale_y_continuous(limits = c(4000, 20000))
final_plot <- p1/p2
final_plot + plot_annotation(
title = 'One-way ANOVA reveals differences in 2022 resale price across different towns',
subtitle = 'Resale price per flat area by Town, 2022',
caption = 'Resale flat princes based on registration date from Jan-2017 onwards from data.gov.sg',
theme = theme(plot.title = element_text(face = "bold", size = 18))
)
Insights
As noted above, the hypothesis testing is done using Kruskal-Wallis test with 95% confidence level for the individual plot. For each plot, they hypothesis is:
H0 : There is no difference between median resale price per flat area across towns
H1 : There is difference between median resale price per flat area across towns
Since the p-value < critical value of 0.05, there is statistical evidence to reject the null hypothesis. We can conclude that for each set of towns (grouped by number of transactions) that there is difference between median resale price per flat area across the towns.
In both plots, it can be observed that there are few pairs of towns with p-value > 0.05 such as Jurong West and Choa Chu Kang. This suggests that the differences between the medians of the pair are not statistically significant.
We can also observe that there is overall higher median and higher number of towns with signficiant outliers (i.e., Central Area, Queenstown, Toa Payoh, Bishan, Clementi) in the second plot (towns with < 850 transactions).
Overall, the findings of the EDA is more or less consistent (refer to section 3.1.2 above).
3.3 Visualising the uncertainty of point estimates
It is tempting to interpret point estimate like median as precise representation of the true data value. However, it is worth noting that there are uncertainties surrounding the point estimates, hence the motivation to display the the target quantile confidence levels (95% or 99%) that the true (unknown) estimate would lie within the interval, given the evidence provided by the observed data.
Design Consideration
ggdist::stat_pointinterval() is used to visualise the uncertainty. Considerations :
The aim is to provide one-glance of levels of uncertainties of median estimate of resale price by town. Hence there is no need to clutter the plot with too much details (i.e., labels and annotations) and it is possible to present this in a single plot
Median point estimates are used instead of mean to adhere with non-parametric testing used above. With this, user can still relate to the above one-way ANOVA analysis
With median used as point estimate, quantile interval is used instead of confidence interval
95% and 99% intervals are used as they are commonly associated with 5% and 1% error rate, which are commonly used in hypothesis testing
Preparation of visualisation
ggplot2 and stat_pointinterval() are used to prepare the plot. Steps taken are
Use the base
ggplot()to determine the data and x, y-variablesDefine confidence levels in
widthargument ofstat_pointinterval(). Define the.pointestimate asmedianand.intervalasqi(quantile intervals). The argumentpoint_coloralso assigns color to the point.scale_color_manual()ofggplot2is used to assign colors for the interval lines.Setting the title, subtitle and caption using
lab()andtheme()ofggplot2
Show the code
#Base ggplot
ggplot(
data = flatpriceclean,
aes(x = town,
y = resale_price_persqm)) +
#Using stat_pointinterval to plot the points and intervals
stat_pointinterval(
aes(interval_color = stat(level)),
.width = c(0.95, 0.99),
.point = median,
.interval = qi,
point_color = "darkred",
show.legend = FALSE) +
#Defining the color of the intervals
scale_color_manual(
values = c("blue2", "darkblue"),
aesthetics = "interval_color") +
#Title, subtitle, and caption
labs(title = 'Visualising uncertainty in median estimates of resale price by town',
subtitle = 'Confidence intervals (95% and 99%) of resale price per flat area by Town, 2022',
caption = 'Resale flat princes based on registration date from Jan-2017 onwards from data.gov.sg',
x = "Planning Town",
y = "Resale Price per sqm (SGD)") +
theme_hc() +
theme(plot.title = element_text(face = "bold", size = 18),
plot.subtitle = element_text(size = 14),
axis.text.x = element_text(angle = 30, hjust = 1))
Insights
As noted above, some towns like Central Area, Bukit Merah, and Bishan have higher uncertainties. This could be potentially due to large presence of outliers in those towns, more varied flat types or models or simply sampling variability problem. On the opposite spectrum, towns like Choa Chu Kang, Sembawang and Bukit Timah have lower uncertainties, which might indicate lower presence of outliers or more uniform flat types.
3.4 Visualisation of Resale Price vs Remaining Lease by Township
Section 3.1 above allow users to perform EDA to explore the relationship between resale price with other variables. Section 3.2 confirms statistical evidence of difference of resale price by towns. It is hence interesting to combine these insights together in one interactive visualization.
Design Consideration
One of the most interesting finding in EDA of two continuous variables is the relationship between resale price per sqm and remaining lease_years. Unfortunately, without applying any filters, there is very little insight can be garnered from this two-dimensional plot. Additional dimensions could be introduced by applying town filter, which was found to be statistically significant in Section 3.2. Considerations:
Interactive bubble plots of resale_price per sqm and remaining_lease_years
Dropdown menu to filter by town. Note that introduction of this filter not only helps users to visualize the different aspects of the bubble plot per town, but also reduces cluttering by too many datapoints.
The bubbles can be colored by flat_type and its size corresponds to the floor_area_sqm. The flat_type is chosen for color as it has the least number of categories, avoiding cluttering the plot with too many colors. The size corresponds to the floor_area_sqm to align with user’s intuition that bigger bubble means bigger size. While the resale price is already normalized against this, users can also see different angle given with remaining_lease_years in the x-axis. Note that with
plot_ly, the color legend can also be selected/deselected to apply additional filterTo reduce cluttering further, opacity is introduced with white border
Tooltip indicating other important variables : resale_price_persqm, floor_area_sqm, remaining_lease_years, flat_model, and storey_range
Preparation of visualisation
Firstly, each unique instance of town was defined using for-loop. This will create a list within list (town_list) containing the method, args, and label arguments of each town. This list will be used as an input to buttons argument of the updatemenus. The annotation variables (annot) for the dropdown menu was also defined.
Refer to code below for more details
town_list <- list()
for (i in 1:length(unique(flatpriceclean$town))) {
town_list[[i]] <- list(method = "restyle",
args = list("transforms[0].value",
unique(flatpriceclean$town)[i]),
label = unique(flatpriceclean$town)[i])
}
annot <- list(list(text = "Planning Town:",
x = 1.41,
y = 0.78,
xref = 'paper',
yref = 'paper',
showarrow = FALSE))plot_ly() is used to prepare the interactive bubble plot. Steps taken are
flatpriceorder variable is created to ensure that there is no clash between the
colorandtransformsargument inplot_ly.Initiating base scatterplot, indicated by
type = 'scatter'.markerargument is used to introduceopacityandline(white plot border). Note thatcolorandsizearguments are specified to beflat_typeandfloor_area_sqmrespectively.Tooltip customisation is created using
hovertemplateargumenttransformargument is set to filter based on selected town provided in theupdatemenuslayoutargument is used to add plot title, x-axis title, and y-axis title. The range of the axis are also fixed to give users fixed reference frame.To create the dropdown menu to select town, the
updatemenusargument is used, passingtown_listin thebuttonsargument. Note thatannotis passed toannotationsargument as well.
Refer to code below for more details
Show the code
flatpriceorder <- flatpriceclean[order(flatpriceclean$flat_type), ]
#Initiating the base plot
plot_ly(data = flatpriceclean,
x = ~remaining_lease_years,
y = ~resale_price_persqm,
type = 'scatter',
mode = 'markers',
size = ~floor_area_sqm,
sizes = c(5, 15),
color = ~factor(flat_type),
marker = list(opacity = 0.6,
sizemode = 'diameter',
line = list(width = 0.2, color = '#FFFFFF')),
#Creating tooltip details
hovertemplate = ~paste("<br>Resale Price per sqm:", resale_price_persqm,
"<br>Floor Area (sqm):", floor_area_sqm,
"<br>Remaining Lease:", remaining_lease_years,
"<br>Flat Model:", flat_model,
"<br>Storey Height:", storey_range),
#Transform plotly by filtering the selected town in updatemenus
transforms = list(list(type = 'filter',
target = ~flatpriceorder$town,
operation = '=',
value = unique(flatpriceorder$town)[1])
)
) |>
#Configuring the plot and axis title and range
layout(title = "Resale Price per flat area increases with remaining lease \nResale transactions by towns, 2022",
xaxis = list(title = "Remaining Lease (Year)",
range = c(40, 100)),
yaxis = list(title = "Resale Price per sqm (SGD)",
range = c(3000, 16000)),
#Creating the dropdown box for users to filter using town
updatemenus = list(list(type = 'dropdown',
xref = "paper",
yref = "paper",
x = 1.4, y = 0.7,
buttons = town_list)
),
annotations = annot
)Insights
This new plot is quite insightful as it allows users to understand the relationship between resale price and remaining lease and flat types by towns/estates. Interesting insights:
The fixed reference frame (x-axis and y-axis range) allows users to see the lower resale price in Woodlands vs higher resale price in Kallang
One can also discern how newer estates like Punggol fetch relatively low resale price (<8kSGD) per sqm despite having > 80 years remaining lease. This is in comparison with Central Area which can fetch up to 14kSGD per sqm for the same remaining lease. This highlights the importance of town centrality in resale price
In a more mature estates like Bedok, Toa Payoh, or Queenstown, the older flats tend to be predominantly dominated by 3-room flats, fetching lower resale price. Between 60-80 years remaining lease, they are dominated by 4 & 5-room flats. Interestingly, beyond 80 years remaining lease, there are significant numbers of 3-room flats sold fetching competitive price.